|Publication number||US7137086 B2|
|Application number||US 10/956,854|
|Publication date||Nov 14, 2006|
|Filing date||Oct 1, 2004|
|Priority date||Jul 6, 2004|
|Also published as||US20060031807, WO2006039637A1, WO2006039637B1|
|Publication number||10956854, 956854, US 7137086 B2, US 7137086B2, US-B2-7137086, US7137086 B2, US7137086B2|
|Original Assignee||Dafca, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (3), Non-Patent Citations (3), Referenced by (10), Classifications (4), Legal Events (6)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is related to U.S. patent application Ser. No. 10/425,101, filed Apr. 28, 2003, U.S. application Ser. No. 10/920,706, filed Aug. 30, 2004, and U.S. Provisional Application No. 60/585,769, filed Jul. 6, 2004 which are hereby incorporated by reference.
This invention relates to integrated circuits and, more particularly, to functional testing and debugging of integrated circuit designs.
Some present day designs of integrated circuits (ICs) comprise a plurality of module, or core, designs (and associated design layouts) that are interconnected within the integrated circuit to create a whole. Such ICs are sometimes referred to as “systems on a chip” (SoCs). The designed SoCs may include core designs that the party designing the integrated circuit has created before, core designs purchased from another party, and core designs that are created specifically for the designed SoC; sometimes is referred to as user defined logic (UDL) block. In connection with the previously created designs, it is recognized that there may exist outputs and/or inputs that are not needed for the specific SoC design undertaken, and those effectively constitute points of access of the core—in addition to the functionally employed inputs and outputs—that may be used for testing the core as well as the entirety of the SoC design.
The aforementioned Ser. No. 10/425,101 patent application discloses a beneficial design approach for SoCs, where each core is encompassed with a wrapper that includes a functionally reconfigurable module (FRM). A wrapper is a collection of elements, including circuit interface elements between essentially every input terminal of the core and circuitry outside the core, and between essentially every output terminal of the core and circuitry outside the core. The FRM can be configurably connected to any of the circuit interfaces, and can also be configured to realize any function. Spare leads are additionally included to permit connectivity from the FRM directly (i.e., not through the circuit interface elements) to circuitry outside the wrapper. The circuitry outside the core can be another wrapper, a UDL block, inputs of the IC, or outputs of the IC.
Each interface element that provides a signal to a core is adapted to configurably provide either an incoming signal (the signal shown in the FIG.) or a signal provided by the associated FRM (not shown in the FIG.). Similarly, each interface element that outputs a signal from a wrapper is adapted to configurably provide either an output of the core (the signal shown in the FIG.) or a signal provided by the associated FRM (not shown in the FIG.).
In is noted that in the illustrative SoC of
In addition to the necessary signal connections between wrappers 11 and 21, and between wrappers 21 and 31, including spare leads, the
In addition, the serial SI input of manger 40 is applied to wrapper 11. The SO output of wrapper 11 is connected to the SI input of wrapper 21. The SO output of wrapper 21 is connected to the SI input of wrapper 31, and the SO output of wrapper 31—which is the Scan-out signal of the
A useful technique for checking the design of integrated circuits (ICs) is called assertion checking. In assertion checking, a collection of conditions is identified that are expected to hold true at any time during the operation of a properly working SoC. The model of the SoC being tested can thus be simulated by application of various input test vectors, and its signals are checked against the collection of assertions. The signals checked by an assertion may include any internal signal. When an assertion “fires,” indicating that the asserted condition that should be met is not met, simulation can stop and the party performing the testing can attempt to analyze the reason for the assertion's failure.
While assertion checking in simulation is a widely used technique, and a simulation run may use hundreds of assertions, employing assertions for debugging hardware is much more constrained.
In general, any assertion can be implemented in hardware, e.g., within the manufactured SoC, but this is a costly approach because it permanently reduces the amount of “real estate” that is available for the normal operation of the SoC. Moreover, in addition to the effective cost associated with appropriated chip “real estate” to implementing hundreds of assertions in the SoC, this solution is not flexible, since the functional problems occurring in silicon often require the implementation of new assertions.
A more economical solution involves using additional hardware to observe many internal signals, run-time programmable logic to select a subset of these, and a run-time programmable assertion logic checking the selected subset of signals, and a clock-control logic to stop the operation of the SoC when an assertion fires. Such a system is illustrated in the paper “Silicon Debug: Scan Chains Alone Are Not Enough” by Gert Jan van Rootselaar and Bart Vermeulen, published in the Proc. International Test Conf., 1999. However, this solution does not provide the desired flexibility, since new signals and new conditions to be checked cannot be added at run-time. As an aside, in this paper the assertions are referred to as “breakpoints.” The described system is an example of at-speed debug, where the tested SoC is exercised in its normal operational mode, with a clock that is at the normal operational speed. The clock is not stopped while the assertions are checked, unless an assertion fires, i.e., fails.
An SoC may have several clock domains, each one having a different system clock signal. Firing of an assertion may stop one or all the functional clocks. In the following, clock stopping may refer to either case.
Usually all the functional registers of an SoC are setup so they can work as scan chains in test mode. Typically, a scan chain has a serial input (SI), a serial output (SO), and a scan-enable control input (SE). When SE is asserted, the register operates as a scan chain, otherwise data are loaded in parallel. (Other scan design techniques exists, but their operation is similar.) SE is typically shared between all registers in the same clock domain. For manufacturing test, the scan chains are separately operated to scan in test data and to scan out the captured data. The same scan chains are also used for debug—when the functional clock has been stopped as a result of an assertion firing (or for any other reason), the internal state of the hardware can be examined (manually or by software) by scanning out all the scan chains. For this operation the test clock shown in
Single-step debug is typically done manually under user control only at several points during the test, because performing scan-out and scan-in operation after every test vector would require too much time. To illustrate, if the test has 100,000 vectors, and the SoC has 500,000 flip-flops, and the test clock runs at 10 MHz, doing a full-scan dump at every cycle would make the entire test run in about two hours. Hence with the current state of the art, one cannot check assertions involving flip-flops on scan-chains, unless these flip-flops are made directly observable (allowing checking through at-speed assertion checking).
An advance in the art is realized by creating SoCs with functionally reconfigurable modules and employing the modules to configure circuitry for performing assertion checking. Since checking assertions with on-chip logic is much faster than checking with external software, the disclosed approach makes possible at-speed assertion checking for at-speed debug. For assertions that involve internal flip-flops on scan chains, the disclosed approach provides the Continuous Single Step (CSS) debug mode, which makes possible automatic checking such assertions after every functional clock.
For an at-speed assertion checking embodiment, each of the cores can be treated as a separate entity with inputs and outputs that can be subjected to assertion checks. The tests are performed in a single FRM that is included in the SoC or, more advantageously, in a number of FRMs that are included in the SoC where each is associated with a particular core of the plurality of cores in the SoC; for example, within a wrapper that encompasses each core. For a CSS assertion checking embodiment, the SI and SO leads of each scan chain of each core are connected to a FRM that is included in the SoC, and a bit extractor/bit injector is implemented in the FRM to efficiently carry out the assertion checking pertaining to each of the associated cores.
As described in the aforementioned Ser. No. 10/425,101 application, an FRM can be realized with field-programmable logic arrays (FPLAs) in a conventional manner, and the variety of digital circuits that can be created within an FPLA is essentially limitless. We realized and discovered that this powerful structure can be applied to provide a novel solution to the problem of efficiently implementing assertion checking in hardware by creating assertions in the reconfigurable logic contained in wrappers. By repeatedly reusing the reconfigurable logic within the FRMs to implement different subsets of assertions, one subset of assertions at a time, all of the necessary assertions can be checked. Advantageously in accord with the principles disclosed herein, all of the assertions of a subset are checked concurrently.
As may be observed from the illustration that the bulk of the assertion checking logic for core 20 is included in FRM 27. One assertion checking module that checks an assertion relative to core 20 but is not included in circuitry 27A is assertion logic module 38, which is implemented within FRM 37. To effect the necessary connection, spare leads between FRM 27 and FRM 37 are employed. The output of module 38 is fed back to FRM 27 to help develop the output of assertion logic module 26-K. As depicted, module 38 also utilizes two functional signals from UDL 30. Because of this, it is absolutely necessary to have some connectivity between FRM 27 and FRM 37. It should be realized, however, that module 38 might need to be implemented within FRM 37 simply because FRM 27 does not have sufficient circuitry for implementing all of the assertion logic modules that are required for the chosen subset of assertions. This can be done, of course, only if there are (a) an adequate number of spare leads that enable the necessary subset of assertion logic modules to implement some of the modules in FRM 73, and (b) FRM 73 has some spare capacity that can be used.
The output developed by each of the assertion logic modules is a logic 0 as long as the assertion holds, and logic 1 the moment the assertion fires (fails). To avoid transients on the output of the assertion logic, the outputs of logic modules 26-i are loaded into flip-flops 29-i at a time when the assertion signals have stabilized. The loading of flip-flops is done by the same system clock that is active in core 20. In
As implied by the above, one might expect that only a subset of all of the assertions that might be desired to test will be checked at any one testing iteration because of the size limitations of the FRM fabric. An SoC that passes all of the assertions that are configured into the FRM, i.e., that passes the configured subset of assertions, can then be re-tested with another subset of the desired assertions.
From the above it should be readily apparent that significant advantages are realized from the almost total flexibility that the
Although, as mentioned above, at-speed assertion checking in accord with the principles disclosed herein is significantly better than, for example, the aforementioned Philips breakpoints checking, it is also appreciated that it does not provide access to the all of the internal variables within each core.
In accord with the principles disclosed herein, the single-step debug approach of the prior art is improved with an on-chip single-step assertion checking approach that makes intimate use of FRMs included within the IC, and which we call Continuous Single Step (CSS) assertion checking.
In CSS, the SoC's testing is divided into intervals that are divided into sub-intervals. Each interval consists of
Note that bits are extracted from the various cores of the SoC concurrently during the second subinterval, that all assertions are checked concurrently during the third subinterval, and that all bit injections are done concurrently during the fourth subinterval. If an assertion fires during the third subinterval, its relevant data (such as the values of the checked signals) are recorded in flip-flops located in a wrapper and these data may be later scanned out of the SoC for further analysis.
It should also be noted that operating in the CSS mode is also very effective because the SoC can be set to any desired state as a starting point for subsequent testing; for example, setting the SoC to the state that it should have been in but for the error that fired an assertion.
To check an assertion that involves values of some flip-flops in the scan chain, in accord with the principles disclosed herein the scan chain is shifted out in the second subinterval. The bits that are needed for testing the subset of assertions that pertain to this scan chain are captured and stored in the shift register 53. The assertions are checked in logic element 27A is the same way as in the
To perform the necessary actions, EXTRACT-INJECT logic module 52 is configured or programmed to recognize when the bits of particular flop-flops appears at the SO output of the scan chain. The particular bits that need to be extracted and captured are known from the subset of assertions that are to be checked. The positions of these desired bits is known from knowledge of the core and how it configures its internal flip-flops into the scan chains, and from knowledge of the subset of assertions that are to be tested. The two sets of knowledge translate to an identification of a sequence of bits that need to be extracted from the sequence of bits that flow out of the SO-1 terminal of the scan chain, for example, the 5th, 28th, . . . , and the 111th bits.
For bit extraction, block 52 sets signal 58 to 0, so that the SO-1 is connected to SI-1 via the multiplexer 55. In this way the first scan chain realizes a circular shift. There are many ways in which bit extractor logic block 52 can be constructed to identify such appearances. One very simple way is to configure an addressable memory that stores the numbers of the identified sequence; in the example above, the memory stores the values 5, 28, . . . 111. Additionally, there is one counter A that advances with each test clock, a counter B that addresses the memory, and a comparator responsive to the output of the memory and to counter A. Thus, when the first entry of the memory is retrieved (e.g., 5), the comparators fires when counter A reaches the value 5, indicating that the fifth bit of the scan chain is available. At that time module 52 sets control signal 54 to the value that enables register 53 to shift, causing it to capture the fifth bit. The same signal also enables counter B to increment, advancing the address and causing the memory to output its second entry, i.e., 28. The process continues until the 28th bit arrives and it is shifted in register 53, etc. After the circular shift of the scan chain is complete, counters A and B are reset to 0, and module 52 sets output signal DONE on lead 57 to 1, indicating that all of the required bits have been extracted and the state of the register has been restored. At this point logic element 27A can perform the asserting checking. [signal 57 should be fed to logic 27A, not to register 53]
Obviously, since the number of flip-flops whose states need to be scanned out is smaller in a single scan chain of a core than in the entire SoC, and since the scan chains of the individual cores are scanned out concurrently, the number of clock ticks that are needed in the second and fourth sub-intervals is basically equal to the length of the longest scan chain, and thus is smaller in the arrangement disclosed herein as compared to prior art single-step debug, where all scan chains are connected serially for a full scan dump. Additionally, because the processing of the extracted values is done in FRMs, there is no need to scan out the raw contents of the internal registers, as done in prior art.
Register 56 is included for bit injection. Bit injection is the means by which the current state of the SoC can be changed. For bit injection, the bits to be injected are stored in the register 56, and the EXTRACT_INJECT control module 52 is configured or programmed to recognize the positions of the bits that need to be replaced. The operation of control module 52 is essentially identical with respect to extracting bits from register 56 and applying them to SI-1 input via multiplexer 55 as its operation relative to register 53. Of course, during the time that the bits are inserted into SI-1, EXTRACT-INJECT control module 52 inverts the control signal 58 that is applied to multiplexer 55.
It is noted that there is no requirement for bit injection to follow the bit extraction of the second subinterval and the assertion checking of the second interval. The bit injection can occur concurrently with the bit extraction. However, it may be desirable to assess the results of the assertion checking and to modify the state of the SoC based on this assessment. Therefore, the bit injection may, at least in some embodiments, take place in the optional fourth subinterval, as described above.
It is also noted that the modification of the current state by a new state represented by the contents of register 56 does not have to follow a failed assertion check. It can simply be a conditional states change that might, upon the mere occurrence of a given state. This would occur without interruption, essentially automatically changing a part of the state transition graph of the finite state machine that is implemented by the associated core.
The FRMs employed in the above description, which are embedded in wrappers, are so depicted for illustrative purposes, and that the FRMs need not be in wrappers. Further, there is no need for creating a number of FRMs that is equal to the number of cores in SoC. Any number of distinct FRMs (though interconnected via “spare leads”) can used.
It should be understood that the principles disclosed herein are described with the aid of a number of illustrative embodiments and that, however, the other and significantly different embodiments can be created that still are within the spirit and scope of this invention. For example, the above describes the assertion checking data acquisition circuit (e.g. elements 53, 55, 56) as being created within the FRMs through a configuration of the FRM. That, however, is not a requirement of the invention, and it is quite possible that at least some of the assertion checking data acquisition circuits might be implemented with fixed circuitry rather than reconfigurable circuitry; perhaps includes in all cores as standard elements.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US6311149 *||Jan 13, 1999||Oct 30, 2001||National Instruments Corporation||Reconfigurable test system|
|US6484280||Sep 30, 1999||Nov 19, 2002||Agilent Technologies Inc.||Scan path test support|
|US6665817||May 5, 2000||Dec 16, 2003||Morphics Technology, Inc.||Apparatus and method for implementing a wireless system-on-a-chip with a reprogrammable tester, debugger, and bus monitor|
|1||D. Kelf and B. Tabbara, "Assertion-Based Verification and Debug in SoC Design," 2003 System-on-Chip Online Design Conf., www.soconline.com.|
|2||G.J. van Rootselaar, B. Vermeulen, "Silicon Debug: Scan Chains Alone Are Not Enough" in Proceedings International Test Conference, Sep. 1999.|
|3||Koranne S: "A Novel Reconfigurable Wrapper for Testing of Embedded Core-Based SOCs and Its Associated Scheduling Algorithm", Journal of Electronic Testing, Kluwer Academic Publishers, vol. 18, No. 4-5, Sep. 2002.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7305635 *||Feb 4, 2005||Dec 4, 2007||Dafca, Inc.||Serial implementation of assertion checking logic circuit|
|US7493247 *||Dec 7, 2005||Feb 17, 2009||Dafca, Inc.||Integrated circuit analysis system and method using model checking|
|US7607057 *||Oct 20, 2009||Lsi Corporation||Test wrapper including integrated scan chain for testing embedded hard macro in an integrated circuit chip|
|US7673206 *||Sep 14, 2007||Mar 2, 2010||Tilera Corporation||Method and system for routing scan chains in an array of processor resources|
|US8302065 *||Mar 13, 2006||Oct 30, 2012||Freescale Semiconductor, Inc.||Device and method for testing a device|
|US20050154947 *||Dec 22, 2004||Jul 14, 2005||International Business Machines Corporation||Microcomputer And Method For Debugging Microcomputer|
|US20060156100 *||Dec 28, 2004||Jul 13, 2006||Boike Mark A||Test wrapper including integrated scan chain for testing embedded hard macro in an integrated circuit chip|
|US20070168730 *||Dec 7, 2005||Jul 19, 2007||Dafca, Inc.||Integrated circuit analysis system and method using model checking|
|US20090077437 *||Sep 14, 2007||Mar 19, 2009||Richard Conlin||Method and system for routing scan chains in an array of processor resources|
|US20090206866 *||Mar 12, 2006||Aug 20, 2009||Freescale Semiconductor, Inc.||Device and method for testing a device|
|Nov 19, 2004||AS||Assignment|
Owner name: DARCA, INC., MASSACHUSETTS
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:ABRAMOVICI, MIRON;REEL/FRAME:016001/0502
Effective date: 20041111
|May 14, 2010||FPAY||Fee payment|
Year of fee payment: 4
|Jun 27, 2014||REMI||Maintenance fee reminder mailed|
|Aug 12, 2014||FPAY||Fee payment|
Year of fee payment: 8
|Aug 12, 2014||SULP||Surcharge for late payment|
Year of fee payment: 7
|Sep 15, 2015||AS||Assignment|
Owner name: SAW VENTURES, LP, TEXAS
Free format text: LICENSE;ASSIGNOR:CEALLAIGH GROUP, LLC;REEL/FRAME:036562/0585
Effective date: 20150705